MAS-Orchestra: Understanding and Improving Multi-Agent Reasoning Through Holistic Orchestration and Controlled Benchmarks

TL;DR

Mas-Orchestra reframes automatic multi-agent design as a training-time, function-calling reinforcement learning problem that enables holistic orchestration across sub-agents. It introduces DoM to control the allowed degree of coordination and MasBench, a five-axis benchmark to study when MAS outperform SAS. Empirical results show MAS gains are task-structure dependent, robust to adversarial inputs, and enhanced when orchestrators are instruction-tuned LLMs rather than purely reasoning models. Together, these contributions offer a principled framework for designing, evaluating, and deploying multi-agent systems with improved reasoning and robustness.

Abstract

While multi-agent systems (MAS) promise elevated intelligence through coordination of agents, current approaches to automatic MAS design under-deliver. Such shortcomings stem from two key factors: (1) methodological complexity - agent orchestration is performed using sequential, code-level execution that limits global system-level holistic reasoning and scales poorly with agent complexity - and (2) efficacy uncertainty - MAS are deployed without understanding if there are tangible benefits compared to single-agent systems (SAS). We propose MAS-Orchestra, a training-time framework that formulates MAS orchestration as a function-calling reinforcement learning problem with holistic orchestration, generating an entire MAS at once. In MAS-Orchestra, complex, goal-oriented sub-agents are abstracted as callable functions, enabling global reasoning over system structure while hiding internal execution details. To rigorously study when and why MAS are beneficial, we introduce MASBENCH, a controlled benchmark that characterizes tasks along five axes: Depth, Horizon, Breadth, Parallel, and Robustness. Our analysis reveals that MAS gains depend critically on task structure, verification protocols, and the capabilities of both orchestrator and sub-agents, rather than holding universally. Guided by these insights, MAS-Orchestra achieves consistent improvements on public benchmarks including mathematical reasoning, multi-hop QA, and search-based QA. Together, MAS-Orchestra and MASBENCH enable better training and understanding of MAS in the pursuit of multi-agent intelligence.

Figures (14)

• Figure 1: Paradigm comparison and Mas-Orchestra. Left: Inference-time orchestration systems typically adopt holistic orchestration, but without training. Mas-Orchestra lies in automatic MAS and formulate the problem as a function-calling RL problem with holistic orchestration. Right: When DoM is configured to be low (dashed lines), the system instantiates at most one agent. When DoM is high, the number of agents is unconstrained. The details of the function-calling protocol and the parser are provided in Appendices \ref{['app:prompt']} and \ref{['app:parser']}.
• Figure 2: Avg@8 (accuracy) for SAS (CoT) and Mas-Orchestra across different axes with Qwen-7b as orchestrator and Qwen-7b and GPT-120b (low) as sub-agent.
• Figure 3: Avg@8 in the Robustness setting with GPT-120b (low) as the sub-agent. (SAS performance is too low to be visible.)
• Figure 4: Avg@8 comparing LLM and RLM as orchestrator.
• Figure 5: Statistics of agents in the generated MAS over the training steps (use Depth equals 4 as an example). The number of agents measures the total number of sub-agents, the sequential agent length measures the length of dependency chain and parallel agent width measures the in degree of sub-agent.